1
00:00:00,510 --> 00:00:08,160
In the last lecture, we discussed how support vector classifier overcomes the limitations of marginal

2
00:00:08,160 --> 00:00:09,390
maximal classifier.

3
00:00:11,040 --> 00:00:18,570
It was by introducing a new parameter called cost and by allowing some amount of misclassification Merza

4
00:00:18,590 --> 00:00:27,540
portrait to classify it still has one limitation support better classify as only work when the observations

5
00:00:27,540 --> 00:00:29,130
are linearly separable.

6
00:00:31,820 --> 00:00:38,630
We have discussed only Linnean hyper plint, but there may be scenarios where the classes are not linearly

7
00:00:38,630 --> 00:00:47,510
separable, such as the scenario four point distributer like this by visual inspection.

8
00:00:47,930 --> 00:00:52,350
We can clearly see that these classes are well separated.

9
00:00:53,510 --> 00:01:01,040
If I can draw a circle here and say that anything within this circle is going to be purple and anything

10
00:01:01,070 --> 00:01:02,450
outside will be blue.

11
00:01:03,200 --> 00:01:06,820
That would be a much better classifier than the one on the right.

12
00:01:10,020 --> 00:01:16,820
So to handle this limitation of linearity, we further generalize and use something known as Cottonelle

13
00:01:16,820 --> 00:01:20,000
method to reach out support vector machines.

14
00:01:24,810 --> 00:01:29,610
Using support vector machines will be able to draw nonlinear boundaries as well.

15
00:01:30,900 --> 00:01:33,570
We'll be discussing support vector machines and the next video.